Re: A client and server encoding question

On Tue, Oct 22, 2013 at 7:00 PM, Albe Laurenz <laurenz.albe@wien.gv.at> wrote:
> Amit Langote wrote:
>> With a server initdb'd with UTF8 encoding , if I create a table with a
>> client using LATIN1 encoding and later try to work with the relation
>> with a client using UTF8 encoding (both the scenarios simulated using
>> single session of psql but with different client_encoding set), there
>> is an error. Following might help illustrate the problem:
>>
>> psql (9.2.4)
>> Type "help" for help.
>>
>> postgres=# SHOW server_encoding;
>>  server_encoding
>> -----------------
>>  UTF8
>> (1 row)
>> Time: 0.761 ms
>>
>> postgres=# SET client_encoding TO LATIN1;
>> SET
>> Time: 1.382 ms
>>
>> postgres=# create table id_äß(ID int);
>> CREATE TABLE
>> Time: 31.344 ms
>>
>> postgres=# \dt
>>         List of relations
>>  Schema |  Name   | Type  | Owner
>> --------+---------+-------+-------
>>  public | id_äß | table | amit
>> (1 row)
>>
>> postgres=# SET client_encoding TO UTF8;
>> SET
>> Time: 1.007 ms
>>
>> postgres=# \dt
>>            List of relations
>>  Schema |     Name     | Type  | Owner
>> --------+--------------+-------+-------
>>  public | id_äÃ\u009F | table | amit
>> (1 row)
>>
>> postgres=# drop table id_äß;
>> ERROR:  table "id_äß" does not exist
>> Time: 1.668 ms
>>
>> postgres=# SET client_encoding TO LATIN1;
>> SET
>> Time: 0.745 ms
>>
>> postgres=# drop table id_äß;
>> DROP TABLE
>> Time: 16.954 ms
>>
>> But, I had an impression that above shouldn't have caused any problem?
>> Should UTF8 handle the situation gracefully? Or am I missing
>> something?
>
> You are missing that your terminal is still running with an UTF8 locale.
>
> So when you create the table, you are feeding psql with \x69645fc3a4c39f:
> 69 ...... "i"
> 64 ...... "d"
> 5f ...... "_"
> c3a4 .... "ä"
> c39f .... "ß"
>
> But you told psql that you are going to feed it LATIN1, so these
> 7 bytes are interpreted as 7 LATIN1 characters, converted to UTF8,
> and the table actually has this name: \x69645fc383c2a4c383c29f
> because the server uses UTF8.
>
> If you change your client encoding back to UTF8, no conversion
> between client and server will take place, and it's hardly
> surprising that the server complains if you tell it to drop
> the table with the name \x69645fc3a4c39f.
>

You are right, I missed the point that my terminal emulator is still
feeding UTF8 into psql.

--
Amit